CN106017454A - Pedestrian navigation device and method based on novel multi-sensor fusion technology - Google Patents

Abstract

The invention discloses a pedestrian navigation device and method based on novel multi-sensor fusion technology. The device includes a handheld smart device platform, an observation processing unit and a fusion filter; the method includes the following steps: (1) The handheld smart device uses its own hardware to acquire The raw data of IMU, magnetometer, pressure gauge, WiFi, BLE and GNSS, etc.; (2) the observation processing unit processes the raw data provided by the handheld smart device to provide observations such as position or velocity to the fusion filter; (3) fusion The filter uses the kinematics model as the system model, and the observation model is established by the result of the observation processing unit, and the pedestrian navigation result is finally obtained through the processing of the fusion filter. The present invention overcomes the shortcoming that navigation errors will accumulate rapidly without other auxiliary systems; the IMU processing module considers multiple modes of hand-held smart devices, and breaks through the traditional multi-sensor fusion IMU needs and carrier fixed limitations; improves Accuracy of pedestrian navigation.

Description

A pedestrian navigation device and method based on novel multi-sensor fusion technology

technical field

The invention relates to the fields of multi-sensor fusion and pedestrian navigation, in particular to a pedestrian navigation device and method based on novel multi-sensor fusion technology.

Background technique

With the development of the mobile Internet, indoor and outdoor pedestrian navigation applications are booming, such as indoor navigation of large shopping malls, patient tracking in hospitals, and crowd flow analysis in supermarkets. Multiple market analysis reports at home and abroad agree that pedestrian navigation will be a research direction with a huge market. At the same time, portable smart devices, such as smartphones, tablets, and smart watches, have been developing at a super-speed in the past decade and have become an indispensable part of people's lives. Most of these portable devices have powerful processors, wireless transceivers, cameras, GNSS receivers and numerous sensors. Therefore, these portable smart devices have become ideal platforms for applications related to multi-sensor fusion and pedestrian navigation.

At present, there are different degrees of defects in single pedestrian navigation technology. Pedestrian navigation technology based on wireless systems such as WiFi and Bluetooth usually has a large fluctuation in wireless signal strength in harsh environments, and cannot provide complete navigation information such as three-dimensional position, velocity, and attitude. System performance is highly dependent on the distribution and number of transmitting devices , position information discontinuous smooth and other defects. Pedestrian navigation technology based on micro-inertial units is accurate in the short term, but navigation errors accumulate quickly. Camera-based visual positioning has disadvantages such as slow calibration of visual sensors, high error rate of feature information extraction, and slow calculation of navigation information in complex environments. Therefore, multi-sensor fusion has become the mainstream solution for pedestrian navigation.

At present, the existing multi-sensor fusion technology generally includes the following steps: (1) Using the measurement data of the inertial unit (three-axis accelerometer and three-axis gyroscope) to calculate the position, velocity and attitude of the tracking object through the inertial mechanical arrangement algorithm (2) Establish the error model corresponding to the inertial mechanical arrangement algorithm, and use it as the system model of the fusion filter; (3) Establish other auxiliary systems (GPS, WiFi, Bluetooth, RFID, GNSS, etc.) as the fusion filter Observation model; (4) Estimate the system state quantity error through the prediction and update process of the fusion filter; and (5) Compensate the system state quantity error for the inertial unit error and the position, velocity and attitude information based on the inertial mechanical arrangement algorithm, to obtain Output the final position information, velocity and attitude information.

The existing multi-sensor fusion technology has the following two fatal shortcomings (1) In the absence of other auxiliary systems, navigation errors will accumulate rapidly; (2) When the inertial unit and the carrier are not fixed, for example: pedestrian navigation For mobile phones and pedestrians, the traditional multi-sensor fusion technology cannot correctly estimate the information of the carrier. Therefore, the existing multi-sensor fusion technology cannot provide accurate pedestrian navigation information in many scenarios.

Contents of the invention

The technical problem to be solved by the present invention is to provide a pedestrian navigation device and method based on novel multi-sensor fusion technology, so as to achieve the effect of improving pedestrian navigation accuracy and usability.

In order to solve the above technical problems, the present invention provides a pedestrian navigation device and method based on a novel multi-sensor fusion technology, including: a handheld smart device platform, an observation processing unit, and a fusion filter; the handheld smart device uses its own hardware to obtain an inertial measurement unit (Inertial Measurement Unit, IMU), magnetometer, pressure gauge, WiFi, Bluetooth Low Energy (Bluetooth Low Energy, BLE) and the raw data of the Global Navigation Satellite System (Global Navigation Satellite System, GNSS), the observation processing unit processes the handheld smart The original data provided by the device provides observations such as position or speed to the fusion filter. The fusion filter uses the kinematics model as the system model, and the observation model is established by the results of the observation processing unit. After processing by the fusion filter, the pedestrian navigation result is finally obtained. .

Preferably, the handheld smart device platform includes common IMUs, magnetometers, pressure gauges, WiFi, low-energy Bluetooth and GNSS etc. in existing smart devices; the IMU provides raw data of acceleration and angular velocity; the magnetometer provides raw data of geomagnetic ;The barometer provides the raw data of atmospheric pressure; WiFi provides the raw data of WiFi Received Signal Strength (RSS); BLE provides the raw data of BLERSS; the GNSS receiver provides the raw data of GNSS; any smart device platform can provide observation information All other sensors can be included in the proposed multi-sensor fusion algorithm.

Preferably, the observation quantity processing unit includes: an IMU processing unit, a magnetometer processing unit, a pressure gauge processing unit, a WiFi processing unit, a BLE processing unit and a GNSS processing unit, etc.; the IMU processing unit processes the original acceleration and angular velocity provided by the IMU data to obtain the IMU position information and send it to the fusion filter; the magnetometer processing unit processes the raw data of the geomagnetic provided by the magnetometer to obtain the geomagnetic position information and sends it to the fusion filter; the manometer processing unit processes the The raw data of atmospheric pressure provided by the manometer is used to obtain elevation information and sent to the fusion filter; the WiFi processing unit processes the RSS raw data provided by the WiFi to obtain WiFi position information and sent to the fusion filter; BLE The processing unit processes the RSS raw data provided by the BLE to obtain BLE position information and transmits it to the fusion filter; the GNSS processing unit processes the position and velocity information provided by the GNSS receiver and transmits it to the fusion filter. The observation processing unit also includes other processing units to process other sensors of the smart device platform to obtain position or velocity information and send it to the fusion filter.

Preferably, the fusion filter includes a system model and an observation model; the system model uses the kinematics model to predict the position and velocity information of the target to be measured, and transmits it to the observation model; the observation model combines the position and velocity information predicted by the system model with the observation The processing unit combines the position and speed information based on IMU, magnetometer, pressure gauge, WiFi, BLE and GNSS to update the final position and speed information of the target to be measured.

Preferably, the IMU processing unit includes a user motion pattern and a device use pattern recognition module, a heading angle deviation estimation module, an improved dead reckoning algorithm module, and a user motion pattern and a device use pattern recognition module according to the IMU of the handheld smart device platform and other The original data provided by optional hardware (such as magnetometer) recognizes user motion patterns such as stationary, walking, running, and equipment usage patterns such as hand-held, short message, telephone, navigation, pocket, backpack; the heading angle deviation estimation module according to the user Motion pattern and equipment usage pattern recognition module output user motion pattern and equipment usage pattern and the raw data provided by the IMU of the handheld smart device platform and other optional hardware (such as a magnetometer) to estimate the heading angle deviation; improve the navigation position The reckoning algorithm module obtains the IMU position information according to the heading angle deviation output by the heading angle deviation estimation module and the IMU of the handheld smart device platform and other optional hardware (such as a magnetometer) and sends it to the fusion filter.

Preferably, the improved dead reckoning algorithm module includes an attitude measurement system module, a heading angle deviation compensation module, a step detection module, a step size estimation module, and a dead reckoning algorithm module, and the attitude measurement system module is based on the IMU and the IMU of the handheld smart device platform. The original data provided by other optional magnetometers can identify the attitude information of the handheld smart device; the heading angle deviation compensation module reads the heading angle deviation output by the heading angle deviation estimation module and compensates the pedestrian heading angle, and outputs it to the dead reckoning algorithm; The detection module algorithm module detects the number of steps of pedestrians according to the raw data of the IMU of the handheld smart device platform and feeds back to the step size estimation module; the step size estimation module is based on the results of the step detection module and the IMU of the handheld smart device platform The raw data estimates the pedestrian's step length, and feeds back to the dead reckoning module; the dead reckoning module outputs the pedestrian heading angle according to the step information output by the step estimation module and the heading angle deviation compensation module The information calculates the IMU position observation and feeds it back to the fusion filter.

Correspondingly, the present invention also provides a pedestrian navigation method based on novel multi-sensor fusion technology, comprising the following steps:

(1) The handheld smart device uses its own hardware to obtain the raw data of IMU, magnetometer, pressure gauge, WiFi, BLE and GNSS; (2) The observation processing unit processes the raw data provided by the handheld smart device to provide observations such as position or speed (3) The fusion filter uses the kinematics model as the system model, and establishes an observation model with the result of the observation processing unit, and finally obtains the pedestrian navigation result through the processing of the fusion filter.

Preferably, the observation processing unit processes the raw data provided by the handheld smart device to provide observations such as position or velocity to the fusion filter, including the following steps:

(1) The IMU processing unit processes the raw data of acceleration and angular velocity provided by the IMU to obtain the IMU position information and sends it to the fusion filter; (2) The magnetometer processing unit processes the raw data of the geomagnetic field provided by the magnetometer To obtain geomagnetic position information and send it to the fusion filter; (3) the manometer processing unit processes the raw data of the atmospheric pressure provided by the manometer to obtain elevation information and send it to the fusion filter; (4) WiFi The processing unit processes the RSS raw data provided by the WiFi to obtain the WiFi position information and transmits it to the fusion filter; (5) the BLE processing unit processes the RSS raw data provided by the BLE to obtain the BLE position information and transmits it to the Fusion filter; (6) GNSS processing unit processes the raw data of GNSS provided by the GNSS chip to obtain position and velocity information of GNSS and transmits to the fusion filter.

Preferably, the IMU processing unit processes the raw data of the acceleration and angular velocity provided by the IMU to obtain the IMU position information and transmits it to the fusion filter including the following steps:

(1) The user motion pattern and device use pattern recognition module recognizes user motion patterns such as stationary, walking, running, and Handheld, SMS, phone, navigation, pocket, backpack and other device usage modes;

(2) The heading angle deviation estimation module provides according to the user motion pattern and the device usage pattern output by the user motion pattern and the device usage pattern recognition module and the IMU and other optional hardware (such as a magnetometer) of the handheld smart device platform The raw data of estimating the heading angle deviation;

(3) The improved dead reckoning algorithm module obtains the IMU position information according to the heading angle deviation output by the heading angle deviation estimation module and the raw data provided by the IMU of the handheld smart device platform and other optional hardware (such as a magnetometer) and sent to the fusion filter.

Preferably, the improved dead reckoning algorithm module includes the following steps:

(1) the attitude measurement system module recognizes the attitude information of the handheld smart device according to the raw data provided by the IMU of the handheld smart device platform and other optional magnetometers;

(2) The course angle deviation compensation module reads the course angle deviation output by the course angle deviation estimation module and compensates the pedestrian course angle, and outputs it to the dead reckoning algorithm;

(3) step detection module algorithm module detects the number of steps of pedestrian according to the raw data of the IMU of described handheld smart device platform and feeds back to the step size estimation module;

(4) the step length estimation module estimates the step length of the pedestrian according to the result of the step detection module and the IMU of the handheld smart device platform, and feeds back to the dead reckoning module;

(5) The dead reckoning module calculates the IMU position observations according to the step size information output by the step size estimation module and the pedestrian heading angle information output by the heading angle deviation compensation module, and feeds back to the fusion filter.

The beneficial effects of the present invention are: the present invention optimizes the use method of the IMU in traditional pedestrian navigation, liberates it from the system model of the fusion filter and becomes an observation model, and overcomes the situation of traditional multi-sensor fusion without other auxiliary systems In this case, the navigation error will accumulate rapidly. The IMU processing module in the present invention considers various modes of handheld smart devices in daily life, and breaks through the limitations of traditional multi-sensor fusion IMU needs and fixed carriers. Therefore, the present invention greatly improves the accuracy and usability of pedestrian navigation.

Description of drawings

Fig. 1 is a schematic structural diagram of a pedestrian navigation device based on a novel multi-sensor fusion of the present invention.

Fig. 2 is a schematic structural diagram of an inertial unit processing module in the present invention.

Fig. 3 is a schematic diagram of a Gaussian kernel support vector machine nonlinear classifier in the present invention.

FIG. 4 is a schematic diagram of a support vector machine for recognition of user motion patterns and smart device usage patterns in the present invention.

Fig. 5 is a schematic diagram of the improved dead reckoning algorithm for pedestrians in the present invention.

detailed description

As shown in Figure 1, a pedestrian navigation device based on new multi-sensor fusion includes: a handheld smart device platform 1, an observation processing unit 2, and a fusion filter 3. The handheld smart device 1 utilizes its own hardware to obtain an inertial measurement unit (Inertial Measurement Unit, IMU) 11, a magnetometer 12, a pressure gauge 13, WiFi 14, Bluetooth Low Energy (Bluetooth Low Energy, BLE) 15 and a global navigation satellite system (Global Navigation Satellite System) SatelliteSystems, GNSS) 16 raw data, the observation processing unit 2 processes the raw data provided by the handheld smart device 1 to provide observations such as position or velocity to the fusion filter 3, and the fusion filter 3 utilizes the kinematics model as the system model 31, An observation model 32 is established based on the results of the observation quantity processing unit, and the pedestrian navigation result is finally obtained through the processing of the fusion filter 3 . The above-mentioned novel multi-sensor fusion pedestrian navigation device can be used in various handheld smart devices (including smart phones, tablet computers, smart watches, etc.), and the handheld smart devices can be held or fixed on pedestrians. The new multi-sensor fusion pedestrian navigation system shown in Figure 1 subverts the use of IMU in traditional pedestrian navigation, liberates it from the fusion filter system model 31 and becomes an observation model 32, overcomes the traditional multi-sensor fusion pedestrian The disadvantage of the navigation device is that the navigation error will accumulate rapidly without other auxiliary systems.

The above-mentioned handheld smart device platform 1 includes IMU 11, magnetometer 12, pressure gauge 13, WiFi 14, BLE 15 and GNSS 16, etc. common to existing smart devices; IMU 11 provides raw data of acceleration and angular velocity; magnetometer 12 provides geomagnetic Raw data; the manometer 13 provides raw data of atmospheric pressure; WiFi 14 provides raw data of WiFi received signal strength (Received Signal Strength, RSS); BLE 15 provides raw data of BLERSS; GNSS 16 provides raw speed and position of GNSS data. Any other sensors of the handheld smart device 1 that can provide observation information can be included in the proposed multi-sensor fusion algorithm.

The above-mentioned observation quantity processing unit 2 includes: an IMU processing unit 21, a magnetometer processing unit 22, a pressure gauge processing unit 23, a WiFi processing unit 24, a BLE processing unit 25, a GNSS processing unit 26, and the like. The IMU processing unit 21 processes the raw data of acceleration and angular velocity provided by the IMU 11 to obtain the IMU position information and transmits it to the fusion filter 3; the magnetometer processing unit 22 processes the raw data of the geomagnetism provided by the magnetometer 12 to obtain Get the geomagnetic position information and send it to the fusion filter 3; the manometer processing unit 23 processes the raw data of the atmospheric pressure provided by the manometer 13 to obtain elevation information and send it to the fusion filter 3; the WiFi processing unit 24 Process the RSS raw data provided by the WiFi 14 to obtain WiFi location information and send it to the fusion filter 3; the BLE processing unit 25 processes the RSS raw data provided by the BLE 15 to obtain BLE location information and send it to the fusion filter Filter 3 ; GNSS processing unit 26 processes the GNSS 16 raw data to obtain image position information and transmits it to the fusion filter 3 . The observation processing unit 2 also includes other processing units to process other sensors of the handheld smart device platform 1 to obtain position or velocity information and send it to the fusion filter 3 .

The aforementioned fusion filter 3 includes a system model 31 and an observation model 32 . The system model 31 uses the kinematics model to predict the position and velocity information of the target to be measured, and transmits it to the observation model 32; Combining the position and speed information of the gauge 12, pressure gauge 13, WiFi 14, BLE 15 and GNSS 16, etc., to update the final position and speed information of the target to be measured.

As shown in FIG. 2 , the IMU processing unit 21 includes a user motion pattern and device usage pattern recognition module 211, a heading angle deviation estimation module 212, and an improved dead reckoning algorithm module 213. The user motion pattern and device usage pattern recognition module 211 is based on the The raw data provided by the IMU 11 of the handheld smart device platform and other optional hardware (such as the magnetometer 12) identify user motion patterns such as stationary, walking, and running, and device usage patterns such as handheld, text message, phone, navigation, pocket, and backpack . The heading angle deviation estimation 212 module is based on the user's motion pattern and the equipment usage pattern output by the user motion pattern and the equipment usage pattern recognition module 211 and the IMU 11 and other optional hardware (such as the magnetometer 12) of the handheld smart device platform Raw data is provided to estimate the heading angle deviation. The improved dead reckoning algorithm module 213 obtains the IMU position information according to the heading angle deviation output by the heading angle deviation estimation module 212 and the raw data provided by the handheld smart device platform 1 IMU 11 and other optional hardware (such as magnetometer 12) and sent to the fusion filter 3. The IMU processing unit in Figure 2 considers the various motion patterns of pedestrians and the various usage modes of smart devices, and designs IMU data processing methods for various usage scenarios, breaking through the limitations of IMU needs and fixed carriers in traditional algorithms , improving the usability of the pedestrian navigation system.

The user motion pattern and device use pattern recognition module 211 uses the existing relevant sensor output of the handheld smart device 1: IMU 11, magnetometer 12, ranging sensor (optional), light sensor (optional). The update frequency of the IMU 11 and the magnetometer 12 is 50-200Hz; the outputs of the latter two sensors are scalar quantities, and the update is triggered by user behavior. User movement patterns and devices use pattern recognition algorithms to extract sensor statistics within 1-3 seconds to make classification decisions. There are many ways to implement the user motion pattern and the device use pattern recognition algorithm. The present invention uses a Gaussian kernel dual support vector machine as an example of implementation.

As shown in Figure 3, the support vector machine based on the Gaussian kernel can implicitly map the feature vector to an infinite-dimensional linear space, so as to achieve or surpass the effect of nonlinear classification (such as traditional KNN). l The prototype of the ¹ -norm soft margin support vector machine is as follows:

Training formula (1):

$\underset{\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; \&xi;</span>}}{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; |</span>{<span\; class="notranslate">\; |</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; C</span>}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; \&xi;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}$

sty _i (w ^T φ(x _i )+b)≥1-ξ _i ,

ξ _i ≥ 0.

Among them, x _i ∈ R ^d , _i ,=1,2,…, are the feature vectors and classification results in the training set, w ∈ R ^d is the weight vector, and C is a controllable normalization constant to balance the data in the training set The overfitting and underfitting of , φ(·) is the feature vector mapping function.

Classification formula (2):

f(x)＝w ^T φ(x)+b.

Since the KKT condition is satisfied, the dual type of ^l1 norm soft margin support vector machine is as follows:

Training formula (3):

$\underset{{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; x</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}\mathrm{<span\; class="notranslate">\; \&phi;</span>}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; T</span>}}\mathrm{<span\; class="notranslate">\; \&phi;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>$

st0≤α _i ≤C,

$\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0.</span>}$

Classification formula (4):

$\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}\mathrm{<span\; class="notranslate">\; \&phi;</span>}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; T</span>}}\mathrm{<span\; class="notranslate">\; \&phi;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; .</span>$

The Gaussian kernel is introduced as an efficient way to calculate mapping and inner product as follows (5):

k(x,x′)＝φ(x) ^T φ(x′)＝exp(-||xx′|| ² /2 ² ),＞0.

Then the dual transformation can be transformed into the following form:

Training formula (6):

$\underset{{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; x</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>$

st0≤α _i ≤C,

$\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0.</span>}$

Classification formula (7):

$\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; .</span>$

Different from the traditional classifier KNN, the optimal solution of support vector machine dual type α _i ,=1,2,…, there are only a few non-zero values, so only a small number of training feature vectors (ie support vectors) need to be retained to participate Online classification calculation (for example, to implement a nonlinear classifier based on randomly generated data as shown in Figure 3, KNN needs to store and use 1000 original training feature vectors, and support vector machines only need 142 support vectors), so that in It greatly reduces the demand for processor battery consumption and system memory, and is more suitable for applications on handheld smart device platforms.

The user movement pattern and device use pattern recognition module 211 classifies pedestrian behavior patterns into five categories: 1. stationary; 2. walking; 3. running; 4. bicycle; 5. driving a car. The identification of pedestrian behavior patterns can be used to apply zero-velocity corrections, adjust the variance of the tracking filter process noise, and tune the correlation time of the Markov process of the dynamical system. The user movement pattern and device use pattern recognition module 211 divides the device use patterns into four categories: 1. Front-end flat; 2. Ear side vertical; 3. Backpack; 4. Armband. The identification of mobile phone attitude patterns can be used to determine the direction of travel (coordinate transformation), and to adjust the variance of tracking filter process noise.

As shown in Figure 4, it is a schematic diagram of a support vector machine for user motion pattern and equipment use pattern recognition using a secondary classifier, including acceleration statistics 2111, angular velocity statistics 2112, rotation angle and tilt angle statistics 2113, light and distance statistics quantity 2114, velocity feedback statistics 2115, feature normalization module 2116, principal component analysis module 2117, support vector machine module 2118, user motion pattern primary classifier 2119 and device usage pattern secondary classifier 2110. The specific implementation steps are as follows: collect representative data sets offline, perform eigenvector normalization and principal component analysis, apply formulas (5) and (6) for training, extract and store support vectors; calculate sensor output statistics online, and conduct Eigenvector normalization and principal component extraction (same as the coefficients in the training set), apply the stored support vectors, and formulas (5) and (7) for secondary classification to determine user motion patterns and device usage patterns.

The heading angle deviation estimation module 212 in the present invention includes many different methods. When only IMU 11 is available, we use a method based on Principal Component Analysis (PCA). A characteristic of pedestrian movement is that the direction of pedestrian acceleration and deceleration is in the direction of travel. Therefore, the traveling direction of the pedestrian can be obtained by analyzing the data of the accelerometer through PCA. When GNSS 16 is available, the direction the pedestrian is traveling can be calculated from the GNSS velocity. When the magnetometer 12 is available, the direction of pedestrian travel can also be calculated by the magnetometer 12 . The heading angle of the handheld smart device is obtained through nine-axis fusion or six-axis fusion. Therefore, the heading angle deviation can be obtained by subtracting the fusion solution of the pedestrian's traveling direction obtained by various methods from the heading angle of the handheld smart device, and output to the heading angle deviation compensation module 2132 .

As shown in Figure 5, it is a schematic diagram of the above-mentioned improved dead reckoning algorithm module 213, including an attitude measurement system module 2131, a course angle deviation compensation module 2132, a step detection module 2133, a step size estimation module 2134, and a dead reckoning algorithm module 2135. The posture system module 2131 recognizes the posture information of the handheld smart device 1 according to the raw data provided by the IMU 11 of the handheld smart device platform 1 and other optional magnetometers 12 . The course angle deviation compensation module 2132 reads the course angle deviation output by the course angle deviation estimation module 212 and compensates the course angle of pedestrians, and outputs it to the dead reckoning algorithm module 2135 . The step detection module 2133 detects the number of steps of pedestrians according to the raw data of the IMU 11 of the handheld smart device platform and feeds it back to the step length estimation module 2134 . The step length estimation module 2134 estimates the pedestrian's step length according to the result of the step detection module and the raw data of the IMU 11 of the handheld smart device platform 1, and feeds back to the dead reckoning module 2135. The dead reckoning module 2135 calculates the IMU position observation according to the step size information output by the step size estimation module 2134 and the pedestrian heading information output by the heading angle deviation compensation module 2132 and outputs it to the fusion filter 3 .

The attitude measurement system module 2131 recognizes the attitude information of the handheld smart device 1 according to the raw data provided by the IMU 11 of the handheld smart device platform 1 and other optional magnetometers 12 . The algorithm adopted by the attitude measurement system module 2131 selects the nine-axis attitude measurement algorithm or the six-axis attitude measurement algorithm according to whether the geomagnetic information is available. Finally, the attitude measurement system module 2131 outputs the heading angle of the smart device 1 to the heading angle deviation compensation module 2132 .

The yaw angle deviation compensation module 2132 reads the yaw angle deviation output by the yaw angle deviation estimation module 212 and

Compensate to the heading angle of the pedestrian, and output to the dead reckoning algorithm module 2135. The specific calculation formula is as follows:

θ _p = θd + θ _offset . (1) where θ _p is the heading angle of the pedestrian, θ _d is the heading angle of the equipment, and θ _offset is the heading angle deviation.

The step detection module 2133 detects the number of steps of pedestrians according to the raw data of the IMU 11 of the handheld smart device platform 1 and feeds it back to the step length estimation module 2134 . Step detection can detect steps by methods such as peak detection, zero-crossing detection, correlation detection and power spectrum detection. The present invention considers multiple user motion patterns and device usage patterns, and the step detection algorithm uses peak detection to simultaneously detect the acceleration data and gyroscope data of the IMU 11 .

The step length estimation module 2134 estimates the pedestrian's step length according to the result of the step detection module 2133 and the raw data of the IMU 11 of the handheld smart device platform 1 , and outputs it to the dead reckoning module 2135 . The step size estimation can be calculated by different methods such as acceleration integration, pendulum model, linear model, empirical model, etc. The present invention takes into account various user motion patterns and device usage patterns, and the step size estimation adopts the following linear model:

s _k-1,k ＝A·(f _k-1 +f _k )+B·(σ _acc,k-1 +σ _acc,k )+C (2)

In the formula, A, B and C are constants, f _k-1 and f _k are the step frequency at k-1 time and k time, σ _acc,k-1 and σ _acc,k are the acceleration at k-1 time and k time Calculated variance.

The dead reckoning module 2135 is based on the position [r _e,k-1 r _n,k-1 ] ^T at time k-1, the step size information s _k-1,k output by the step size estimation module 2134 and the heading angle deviation compensation module The heading angle information θ _k-1 output by 2132 calculates the position [r _e,k r _n,k ] ^T at time k. The corresponding calculation formula is as follows:

$\left[\begin{array}{c}{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}}\\ {\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}}\end{array}\right]<span\; class="notranslate">\; =</span>\left[\begin{array}{c}{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}\\ {\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}\end{array}\right]<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>\left[\begin{array}{c}{\mathrm{<span\; class="notranslate">\; cos\&theta;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}\\ {\mathrm{<span\; class="notranslate">\; sin\&theta;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}\end{array}\right]<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

Finally, the dead reckoning module 2135 outputs the IMU position observations to the fusion filter 3 .

The fusion filter 3 includes a system model 31 and an observation model 32 . The traditional multi-sensor fusion structure generally processes the IMU measurement data through the inertial machine orchestration algorithm, and establishes the relevant fusion filter system model. Due to the many integral operations in the inertial mechanical orchestration, the positioning errors of conventional multi-sensor fusion architectures accumulate rapidly when there is no extrinsic assistance system. The invention overcomes the defects of the traditional multi-sensor fusion structure, uses the motion model of pedestrians as the system model, and uses IMU related data as the observation model like other systems.

The fusion filter 3 may use a Kalman filter (Kalman Filter, KF), an adaptive Kalman filter (Adaptive Kalman Filter, AKF), a UKF lossless Kalman filter (Unscented Kalman Filter, UKF) or a particle filter (Particle Filter, PF). The present invention gives a design example of KF. Other filters can refer to the design of KF. Fusion filter 3 is implemented as the state vector of KF defined as follows:

x＝[r _e r _n r _u v _e v _n v _u ] ^T (4)

In the formula, re, rn and ru are the three-dimensional positions (northeast sky coordinate system), and ve, vn and vu are the corresponding three-dimensional velocity components. The KF system model 32 adopts a classical kinematics model, which is defined as follows:

x _k+1|k ＝Φ _k,k+1 x _k|k +ω _k (5)

where x _k+1|k is the predicted state vector, x _k|k is the previous state vector at time k, Φ _k,k+1 is a 6×6 turn

Shift matrix:

${\mathrm{<span\; class="notranslate">\; \&Phi;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>\left[\begin{array}{cccccc}\mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; \&Delta;</span>}\mathrm{<span\; class="notranslate">\; t</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; \&Delta;</span>}\mathrm{<span\; class="notranslate">\; t</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; \&Delta;</span>}\mathrm{<span\; class="notranslate">\; t</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}\end{array}\right]<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 6</span>}<span\; class="notranslate">\; )</span>$

where Δt is the time difference between two moments. ωk is the covariance _matrix The processing noise of is defined as follows:

In the formula and is the velocity noise in the east sky coordinate system at time k, modeled as a random walk.

In the formula and is the velocity noise of the east sky coordinate system at time k-1, _ne , n _n and n _u are Gaussian white noises, and Δt is the time difference between two moments.

The fusion filter 3 is implemented as a KF measurement model 31 defined as follows:

z _k ＝H _k x _k|k +υ _k (9)

where z _k is the measurement vector, and H _k is the decision matrix. υ _k is the measurement noise modeled with Gaussian white noise, and its covariance matrix is z _k and H _k vary with different observations. When the observations come from the IMU 11, typical z _k and H _k are defined as follows:

z _k ＝[r _e r _n ] ^T

${\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; =</span>\left[\begin{array}{cccccc}\mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\end{array}\right]<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; )</span>$

When the observations come from the magnetometer 12, typical z _k and H _k are defined as follows:

z _k ＝[r _e r _n r _u ] ^T

${\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; =</span>\left[\begin{array}{cccccc}\mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\end{array}\right]<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 11</span>}<span\; class="notranslate">\; )</span>$

When the observed quantity comes from the manometer 13, typical z _k and H _k are defined as follows:

z _k = r _u

H _k ＝[0 0 1 0 0 0] (12) When the observations come from WiFi 14, typical z _k and H _k are defined as follows:

z _k ＝[r _e r _n r _u ] ^T

${\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; =</span>\left[\begin{array}{cccccc}\mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\end{array}\right]<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 13</span>}<span\; class="notranslate">\; )</span>$

When the observations come from BLE15, typical z _k and H _k are defined as follows:

z _k ＝[r _e r _n r _u ] ^T

${\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; =</span>\left[\begin{array}{cccccc}\mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\end{array}\right]<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 14</span>}<span\; class="notranslate">\; )</span>$

When the observations come from GNSS16, typical z _k and H _k are defined as follows:

z _k ＝[r _e r _n r _u v _e v _n v _u ] ^T

${\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; =</span>\left[\begin{array}{cccccc}\mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; 1</span>}\end{array}\right]<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 15</span>}<span\; class="notranslate">\; )</span>$

The KF process has two phases: prediction and update. In the prediction process, the state vector and covariance matrix are predicted according to the system model.

$\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&Phi;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; k</span>}}\\ {\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&Phi;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; \&Phi;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 16</span>}<span\; class="notranslate">\; )</span>$

During the update, the state vector and covariance matrix are updated according to the measured model:

$\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span>}^{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}\\ {\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; )</span>\\ {\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 17</span>}<span\; class="notranslate">\; )</span>$

In the formula, K _k is called the Kalman gain.

Although the present invention has been illustrated and described in terms of preferred embodiments, those skilled in the art should understand that various changes and modifications can be made to the present invention without departing from the scope defined by the claims of the present invention.

Claims (12)

1. A pedestrian navigation device based on novel multi-sensor fusion technology, characterized in that, comprising: a handheld smart device (1), an observation processing unit (2) and a fusion filter (3); the handheld smart device ( 1) Utilize its own hardware to obtain raw observational data, the observational processing unit (2) processes the raw data provided by the handheld smart device (1) to provide observations such as position or velocity to the fusion filter (3), the fusion The filter uses the kinematics model as the system model (32), and the observation model (33) is established from the result of the observation processing unit, and the pedestrian navigation result is finally obtained through the processing of the fusion filter (3). 2. The pedestrian navigation device based on novel multi-sensor fusion technology as claimed in claim 1, characterized in that, said handheld smart device 1 own hardware includes: inertial measurement unit IMU (11), magnetometer (12), pressure gauge (13), WiFi (14), Bluetooth Low Energy Consumption BLE (15) and Global Navigation Satellite System GNSS (16); Described IMU (11) provides the raw data of acceleration and angular velocity; Described magnetometer (12) provides geomagnetism The raw data of the manometer (13) provides the raw data of the atmospheric pressure; the WiFi (14) provides the raw data of the WiFi received signal strength RSS; the BLE (15) provides the raw data of the BLERSS; the GNSS receives The machine (16) provides GNSS raw data; the handheld smart device (1) can also include other sensors that provide observation information. 3. the pedestrian navigation device based on novel multi-sensor fusion technology as claimed in claim 1, is characterized in that, observation quantity processing unit module (2) comprises: IMU processing unit (21), magnetometer processing unit (22), pressure computer processing unit (23), WiFi processing unit (24), BLE processing unit (25) and GNSS processing unit (26); the IMU processing unit (21) processes the original acceleration and angular velocity provided by the IMU (11) data to obtain the IMU position information and send to the fusion filter (3); the magnetometer processing unit (22) processes the raw data of the geomagnetism provided by the magnetometer (12) to obtain the geomagnetic position information and send to the The fusion filter (3); the manometer processing unit (23) processes the raw data of the barometric pressure provided by the manometer (13) to obtain elevation information and transmits it to the fusion filter (3); the The WiFi processing unit processes (24) the RSS raw data provided by the WiFi (14) to obtain the WiFi location information and sends it to the fusion filter (3); the BLE processing unit processes (25) the BLE (15) Provided RSS raw data to obtain BLE position information and transmit to the fusion filter (3); the GNSS processing unit (26) processes the position and velocity information provided by the GNSS receiver (16) and transmits to the Fusion filter (3); observation processing unit module (2) also includes other processing units to process other sensors of the smart device platform to obtain position or velocity information and send it to the fusion filter (3). 4. the pedestrian navigation device based on novel multi-sensor fusion technology as claimed in claim 1, is characterized in that, described IMU processing unit (21) comprises user motion pattern and equipment use pattern recognition module, heading angle deviation estimation module, improved The dead reckoning algorithm module, the user motion pattern and device use pattern recognition module recognizes user motion patterns such as stationary, walking, running, and handheld, Short message, phone, navigation, pocket, backpack and other equipment usage patterns; the user motion pattern and equipment usage pattern output by the user motion pattern and equipment usage pattern recognition module according to the user motion pattern and the equipment usage pattern of the described heading angle deviation estimation module and the information of the handheld smart device platform The raw data provided by the IMU and other optional hardware estimates the heading angle deviation; the improved dead reckoning algorithm module outputs the heading angle deviation according to the heading angle deviation estimation module and the IMU of the handheld smart device platform and other optional The raw data provided by the hardware obtains the IMU position information and transmits it to the fusion filter. 5. the pedestrian navigation device based on novel multi-sensor fusion technology as claimed in claim 4, is characterized in that, described improved dead reckoning algorithm module comprises attitude measurement system module, heading angle deviation compensation module, step detection module, step length Estimation module, dead reckoning algorithm module, the attitude measurement system module recognizes the attitude information of the handheld smart device according to the raw data provided by the IMU of the handheld smart device platform and other optional magnetometers; the heading angle deviation compensation module Read the heading angle deviation output by the heading angle deviation estimation module and compensate the pedestrian heading angle and output to the dead reckoning algorithm; the step detection module algorithm module detects the pedestrian's step feedback according to the original data of the IMU of the handheld smart device platform Give the step-length estimation module; The step-length estimation module estimates the step-length of the pedestrian according to the result of the step detection module and the IMU of the handheld smart device platform, and feeds back to the dead reckoning module; The calculation module calculates the IMU position observation according to the step size information output by the step size estimation module and the pedestrian heading angle information output by the heading angle deviation compensation module, and feeds it back to the fusion filter. 6. The pedestrian navigation device based on novel multi-sensor fusion technology as claimed in claim 4, wherein the user movement pattern and equipment use pattern recognition module adopts a two-level classification based on Gaussian kernel dual support vector machine The sensor extracts the sensor output of handheld smart devices in periodic time intervals, and uses support vectors and Gaussian kernels to identify and classify user motion patterns and mobile phone usage patterns. 7. The pedestrian navigation device based on novel multi-sensor fusion technology as claimed in claim 1, characterized in that: calculate the output statistics of IMU, magnetometer, optional ranging, optional optical sensor, and derived statistics , using it as the eigenvector for normalization and principal component analysis, intercepting the eigenvector corresponding to the eigenvalue that retains most of the data variance to participate in the offline training of the support vector machine. 8. The pedestrian navigation device based on new multi-sensor fusion technology as claimed in claim 1, characterized in that: the online classification calculation of the handheld smart device platform only needs to store and use a small number of support vectors after dimensionality reduction to perform user motion patterns Secondary classification related to device usage patterns. 9. A pedestrian navigation method based on novel multi-sensor fusion technology, comprising the steps of: (1) The handheld smart device uses its own hardware to obtain the raw data of IMU, magnetometer, pressure gauge, WiFi, BLE and GNSS; (2) The observation processing unit processes the raw data provided by the handheld smart device to provide observations such as position or velocity to the fusion filter; (3) The fusion filter uses the kinematics model as the system model, and establishes the observation model with the result of the observation processing unit, and finally obtains the pedestrian navigation result through the processing of the fusion filter. 10. A pedestrian navigation method based on novel multi-sensor fusion technology as claimed in claim 9, wherein the observation processing unit processes the raw data provided by the handheld smart device to provide observations such as position or velocity to the fusion filter Including the following steps: The IMU processing unit processes the raw data of acceleration and angular velocity provided by the IMU to obtain IMU position information and transmits it to the fusion filter; The magnetometer processing unit processes the raw geomagnetic data provided by the magnetometer to obtain geomagnetic position information and transmits it to the fusion filter; The manometer processing unit processes the raw data of atmospheric pressure provided by the manometer to obtain elevation information and transmits it to the fusion filter; The WiFi processing unit processes the RSS raw data provided by the WiFi to obtain WiFi location information and transmits it to the fusion filter; The BLE processing unit processes the RSS raw data provided by the BLE to obtain the BLE position information and transmits it to the fusion filter; The GNSS processing unit processes the GNSS raw data provided by the GNSS chip to obtain GNSS position and velocity information and transmits them to the fusion filter. 11. A kind of pedestrian navigation method based on novel multi-sensor fusion technology as claimed in claim 10, it is characterized in that, IMU processing unit processes the raw data of acceleration and angular velocity that IMU provides to obtain IMU position information and send to said fusion The filter consists of the following steps: According to the original data provided by the IMU of the handheld smart device platform and other optional hardware, the user motion pattern and device use pattern recognition module recognizes user motion patterns such as stillness, walking, running, and hand-held, short message, phone, navigation, etc. Device usage modes such as pockets and backpacks; The heading angle deviation estimation module estimates the heading according to the user motion pattern and device usage pattern output by the user motion pattern and device usage pattern recognition module and the raw data provided by the IMU of the handheld smart device platform and other optional hardware angular deviation; The improved dead reckoning algorithm module obtains the IMU position information according to the heading angle deviation output by the heading angle deviation estimation module and the IMU of the handheld smart device platform and other optional hardware and sends it to the fusion filter . 12. A kind of pedestrian navigation method based on novel multi-sensor fusion technology as claimed in claim 11, is characterized in that, improved dead reckoning algorithm module comprises the steps: The attitude measurement system module recognizes the attitude information of the handheld smart device according to the raw data provided by the IMU of the handheld smart device platform and other optional magnetometers; The heading angle deviation compensation module reads the heading angle deviation output by the heading angle deviation estimation module and compensates the heading angle of the pedestrian, and outputs it to the dead reckoning algorithm; The step detection module algorithm module detects the number of steps of the pedestrian according to the raw data of the IMU of the handheld smart device platform and feeds back to the step size estimation module; The step length estimation module estimates the step length of the pedestrian according to the result of the step detection module and the IMU raw data of the handheld smart device platform, and feeds back to the dead reckoning module; The dead reckoning module calculates the IMU position observations according to the step size information output by the step size estimation module and the pedestrian heading angle information output by the heading angle deviation compensation module, and feeds back to the fusion filter.